ESS101-Modeling and Simulation: Paolo Falcone

Department of Signals and Systems
ESS101- Modeling and Simulation

Lecture 7
Paolo Falcone
Chalmers University of Technology
Gteborg, Sweden
Paolo Falcone (CHALMERS)
ESS101 Modeling and Simulation
September 9, 2016
Physical modelling
September 9, 2016
Physical modeling. Three phases method
Structuring
Divide into subsystems
Inputs, outputs, internal variables?
Graph or block diagram
Relationships
Conservation laws
Constitutive relations
differential equations
and algebraic relationships
Form state-space model

Choose state variables
Rearrange the equations
x = Ax + Bu
y = Cx + Du
September 9, 2016
System identification
Limitations in physical modeling:

Constitutive relationships may be unknown
Physical parameters may be either unknown or
highly uncertain
The problem is too complex
Basic idea in SysId
System
Collect measurements of u and y and find a

model for the System fitting the collected data
September 9, 2016
System identification
w(t) input
disturbance
u(t)
System
y(t)
ey(t) measurement
noise
eu(t) measurement
noise
u(kT)
y(kT)
Model derived from input and output

measurements
In practice, measurements are noisy (eu, ey) and
system is affected by disturbances
Goal. Model the effects of disturbances
September 9, 2016
Today (Chapter 3)
Disturbance modeling
Deterministic models in time domain
Stochastic models in time domain
Deterministic models in the frequency domain
Stochastic models in the frequency domain
September 9, 2016
Time domain deterministic models
IDEA: to describe the disturbance signal as the output of a

dynamical model
x! w (t ) = f (xw (t ), uw (t ) )
w(t ) = g (xw (t ), uw (t ) )
or
W ( s) = Gw (s)U w ( s)
if f and g are linear
where typical choices for uw are pulse, pulse train or

sinusoids. In discrete time
w(k + n) + d1w(k + n 1) + ! + d n w(k ) =
c0uw (k + n) + c1uw (k + n 1) + ! + cnuw (k )
or
c0 z n + c1 z n 1 + ! + cn
Gw ( z ) = n
z + d1 z n 1 + ! + d n
Examples in
the textbook
September 9, 2016
Time domain stochastic models
September 9, 2016
We describe a signal through a stochastic model when we are not

able to predict it. E.g., the wind gust velocity.
Although such a signal cannot be predicted, a guess can be made on
its expected value, based on its stochastic characterizations
A signal x(t) can be viewed as a Random Variable (RV). As such, x(t)

can be characterized by means of its Cumulative Distribution Function
(CDF)
Fx (x,t) = Pr{x(t) x}
September 9, 2016

CDF can be interpreted as
N(x,t)
Fx (x,t) = lim
N
N
where N(x,t) is the number of times x(t) is

below the threshold x over N realizations
(tosses)
For the RV x(t), we also define the Probability Density Function

(PDF) as
Fx (x,t)
f x (x,t) =
x
Its integral, over a range of x, is the probability the value of the RV
falls in that range
September 9, 2016
10
A RV can be more compactly described by mean of the expected

value or mean
mx (t) = E[x(t)] =
xf x (x, t)dx
the covariance
Rx (t,s) = E ( x(t) mx (t))( x(s) mx (s))

the cross covariance
Rxw (t) = E ( x(t) mx (t))( w(t) mw (t))
zero for independent random variables

the variance
Vx (t) = Rx (t,t)
September 9, 2016
11
Stationary signals
A signal x(t) is stationary if m x (t) = m x

the covariance Rx (t,s) = Rx (t s) = Rx ( )
the cross covariance Rxw (t) = Rxw
the variance Vx (t) = Vx
That is, the signal x(t) is stationary if

its stochastic
characterizations do not depend on the time
September 9, 2016
12
Special signals. White noise
A white noise e(t) is a sequence of independent, normally distributed

stochastic variables
f (x) =
1
2
(x ) 2
e(t) N (0, )
2 2
=2
Mean Variance
September 9, 2016
13
White noise
White Noise. Used to describe disturbances and noises in

engineering systems. A sequence of independent, normally
distributed random variables
( me (t) = me = 0
e(t) N(0, ) )
* Re (t,s) = Re (t s) = Re ( ) = ( )
Colored Noise. Obtained from white noise by filtering it through
linear systems.
September 9, 2016
14
Stochastic processes. Exercise.
e(t ) N (0,1)
e(t )
1
H ( z) =
1 0.9 z 1
w(t )
Compute mean and covariance of w(t).

Watch out the covariance!!
September 9, 2016
15
Exercise
September 9, 2016
16
Covariance
September 9, 2016
18
Frequency domain models
September 9, 2016
19
(Density) Spectrum
w ( ) describes the frequency content of the signal w(t)

2
w ( )d
is a measure of the signal energy in the frequency
interval 1 , 2
September 9, 2016
20
Spectrum. Continuous deterministic signals

Consider the signal w(t ), < t < with
w(t ) dt <
The spectrum is defined as:

w ( ) = W ()
where
W ( ) = w(t )e it dt
(Fourier transform)
Recall that
w(t) =
a e
n
n=
in t
with
1
an =
T
T
2
T
w(t)e
!
2
i nt
T
September 9, 2016
dt
21
Spectrum. Sampled deterministic signals

Consider the signal w(k ), < t < with
w(kT ) <
k =
The spectrum is defined as:

T
w
( ) = W ( )
where
W T ( ) = T w(k)e ikT
(Fourier transform)
k =
September 9, 2016
22
Spectrum. Sampled stochastic processes
The spectrum for a sampled stochastic signal w(k) is

defined as
k =
Tw ( ) = T Rw ( kT )e ikT
k =
The cross spectrum of two signals u and y is defined as
k =
Tyu ( ) = T Ryu ( kT )e ikT

k =
and provide information their joint variation
September 9, 2016
23
Examples
f = 200Hz
f = 100Hz
x(t) = cos(2ft) + n(t)

f = 300Hz
f = 400Hz
September 9, 2016
24
Example
Lets calculate the spectrum of a white noise w(t) with

variance
Use the definition of spectrum for stochastic signals
k =
Tw ( ) = T Rw ( kT )e ikT
k =
Recall that Re (t) = ( t )
Hence
Tw ( ) = T
September 9, 2016
25
Spectra and linear systems
Consider the sampled signal y(k) obtained as output of

a linear system
y(k) = G(q)u(k) + v(k)
where u(k) and v(k) are uncorrelated

Ty ( ) and Tyu ( ) are given by
The spectra
T
y
( ) = G(i ) Tu ( ) + Tv ( )
Tyu ( ) = G(i )Tu ( )
September 9, 2016
26
Spectra and linear systems. Example
Let w(t) be a stationary stochastic process. w(t) is

generated from the time discrete relation
w(t ) 0.4w(t 1) = e(t ) 0.5e(t 1)

where e(t) is white noise with variance . Determine the
spectrum of w(t)
September 9, 2016
27
Example. Vehicle Dynamics

Ftractive
Fwind
mv! = Ftractive Froll Fwind Fgrade
with
Fgrade
mg
Froll
Froll = kmgsign (v)

1
C x A(v vwind )2
2
= mg sin
Fwind =
Fgrade
The road grade might be available (e.g., through GPS and maps).
Wind velocity is not
September 9, 2016
29
Road slope spectra
September 9, 2016
30

ESS101-Modeling and Simulation: Paolo Falcone

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ESS101-Modeling and Simulation: Paolo Falcone

Uploaded by

Copyright:

Available Formats

Department of Signals and Systems